Apple Global · Production
Every request passes the gate
1,284 requests in the last hour. Three were blocked, twenty-two masked, and four tool calls are held for a person.
Where things stand
Apple Global · Production
1,284 requests in the last hour. Three were blocked, twenty-two masked, and four tool calls are held for a person.
Where things stand
Everything registered in this organisation, and what each one is doing right now.
| Agent | Status | Environment | Model | Runs 24h | p95 | Cost 30d |
|---|---|---|---|---|---|---|
| RAResearch Agentresearch-agent-04 | Running | Production | Claude Sonnet 4.6 | 318 | 2.4s | ₹8,400 |
| STSupport Triagesupport-triage-11 | Running | Production | Claude Haiku 4.5 | 1,204 | 840ms | ₹3,120 |
| FAFinOps Advisorfinops-advisor-02 | Degraded | Staging | Claude Sonnet 4.6 | 86 | 6.1s | ₹940 |
| SSSecOps Sentinelsecops-sentinel-03 | Running | Production | Claude Sonnet 4.6 | 452 | 1.9s | ₹5,260 |
| DDDeal Desk Reviewerdeal-desk-07 | Stopped | Development | Claude Haiku 4.5 | 0 | — | ₹0 |
| KRK8s Remediationk8s-remediation-01 | Awaiting approval | Staging | Claude Sonnet 4.6 | 12 | 3.3s | ₹410 |
Select a row to open the inspector. Updated 12 seconds ago.
Runs, latency and spend across every agent this organisation operates.
Two plots, one shared time axis — runs and latency have different units, so they never share a scale.
Recent runs
| Time | Agent | Outcome | Trigger | Duration | Tokens | Cost |
|---|---|---|---|---|---|---|
| 10:42:15 | Research Agent | Completed | Chat | 2.4s | 18,420 | ₹26 |
| 10:42:08 | Support Triage | Completed | Schedule | 0.9s | 3,180 | ₹4 |
| 10:41:55 | FinOps Advisor | Refused | Schedule | 0.4s | 610 | ₹1 |
| 10:41:30 | SecOps Sentinel | Completed | Trigger | 1.9s | 9,240 | ₹13 |
| 10:41:12 | K8s Remediation | Awaiting approval | Trigger | — | 2,050 | ₹3 |
| 10:40:47 | Research Agent | Failed | Chat | 11.2s | 24,110 | ₹34 |
What it does
Researches counterparties across public filings, internal documents and the web, then produces a sourced brief for the deal desk.
Connected tools
Current deployment
| Run | Started | Outcome | Trigger | Duration | Tokens | Cost |
|---|---|---|---|---|---|---|
| run_8f21c4 | 10:42:15 | Completed | Chat | 2.4s | 18,420 | ₹26 |
| run_8f219e | 10:38:02 | Completed | Chat | 3.1s | 21,050 | ₹29 |
| run_8f2188 | 10:31:44 | Failed | Schedule | 11.2s | 24,110 | ₹34 |
| run_8f2170 | 10:24:19 | Completed | Chat | 1.8s | 12,880 | ₹18 |
| run_8f215b | 10:19:55 | Refused | Chat | 0.3s | 480 | ₹1 |
| Version | Phase | Released | By | Gate | Replicas |
|---|---|---|---|---|---|
| v2.4.2 | Awaiting approval | — | F. Manzer | Tests passed | — |
| v2.4.1 | Production | 11 May, 09:24 | F. Manzer | Tests passed | 4 / 4 |
| v2.4.0 | Staging | 08 May, 16:02 | A. Rahman | Tests passed | 1 / 1 |
| v2.3.7 | Retired | 28 Apr, 11:40 | A. Rahman | 2 failures | — |
The chat transcript lives here — threads on the left, the conversation in the middle, and each step of the agent's reasoning expandable inline. Not built in this preview.
Scheduled triggers, their run history and the approval queue for this agent. Not built in this preview.
Attached MCP servers and capability grants, with what each one currently yields and why any tool is excluded. Not built in this preview.
Model, iteration limits, guardrail assignments, owning team and region scope. Not built in this preview.
Authoring: agent definitions, their test suites, and the tools they can be granted.
The left panel shows how this section subdivides. Its screens aren't part of this preview — the point here is that every section carries its own contextual panel, and the top bar stays put.
This section only adds things. When a connection is saved its credential goes to Secrets and the server itself appears in Inventory — which is where you manage it from then on.
Which models this organisation may use, what they cost, and which one answers when.
| Model | Provider | Source | Context | In / 1M | Out / 1M | Used by |
|---|---|---|---|---|---|---|
| Claude Sonnet 4.6claude-sonnet-4-6 | Anthropic | Platform default | 200k | $3.00 | $15.00 | 11 agents |
| Claude Haiku 4.5claude-haiku-4-5 | Anthropic | Platform default | 200k | $0.80 | $4.00 | 6 agents |
| Claude Opus 5claude-opus-5 | Anthropic | Added by org | 200k | $15.00 | $75.00 | 1 agent |
| GPT-4ogpt-4o | OpenAI | Platform default | 128k | $2.50 | $10.00 | 2 agents |
| Gemini 2.5 Progemini-2.5-pro | Added by org | 1M | $1.25 | $10.00 | — | |
| Sarvam-Msarvam-m | Sarvam | Added by org | 32k | $0.10 | $0.30 | — |
Platform defaults come from the deployment’s own configuration and cannot be edited here. Models added by the organisation carry their own provider key.
Every credential this organisation holds, wherever it was entered. Secrets are sealed in an encrypted vault, and a value is never returned to the browser — not to this page, not to an administrator, not through the API. A secret can be rotated, revoked or deleted. It can never be read back.
| Secret | Category | Used by | Added by | Last rotated | Status |
|---|---|---|---|---|---|
| github-mcp-token | MCP server | GitHub MCP | Farzan Manzer | 12 May | Active |
| postgres-mcp-dsn | MCP server | PostgreSQL MCP | Arnav | 04 May | Active |
| slack-bot-token | MCP server | Slack MCP | Amit | 22 Apr | Rotate soon |
| filesystem-mcp-key | MCP server | Filesystem MCP | Farzan Manzer | 09 May | Active |
| dealdesk-a2a-key | External agent | Deal Desk Reviewer | Arnav | 09 May | Active |
| anthropic-api-key | Model provider | Sonnet 4.6, Haiku 4.5, Opus 5 | Platform | 01 May | Active |
| openai-api-key | Model provider | GPT-4o | Farzan Manzer | 28 Apr | Active |
| sarvam-api-key | Model provider | Sarvam-M | Farzan Manzer | 19 Mar | Rotate soon |
| tavily-search-key | Tool | Tavily search | Platform | 30 Apr | Active |
| smtp-credential | Organisation | Invites, verification mail | Farzan Manzer | 08 Feb | Last test failed |
| gateway-log-secret | Organisation | Gateway log collector | Platform | 02 May | Active |
Select a secret to rotate, revoke or delete it. Secrets are added in the flow that needs them — connecting a server, adding a model provider, configuring mail — so this page is the register, not another place to type one in.
Rules that apply to agents — where each one comes from, and which wins when two disagree. A policy is written once and bound to a layer; the layer decides its reach.
| Policy | Category | Jurisdiction | Rules | Bound at | Revision | Status |
|---|---|---|---|---|---|---|
| PII redaction | Data protection | Global | 6 | Organisation | r4 | Active |
| Prompt-injection screening | Safety | Global | 5 | Organisation | r3 | Active |
| GDPR data minimisation | Data protection | EU · GDPR | 7 | Organisation | r3 | Active |
| EU AI Act transparency | Compliance | EU · AI Act | 4 | Organisation | r2 | Active |
| India DPDP consent | Data protection | India · DPDP Act | 5 | Sub-org · Engineering | r1 | Active |
| Australian Privacy Principles | Data protection | Australia · Privacy Act | 6 | Organisation | — | Draft |
| HIPAA safe harbour | Data protection | United States · HIPAA | 8 | Organisation | — | Draft |
| No customer data to tools | Data protection | Global | 3 | Organisation | r7 | Active |
| Production approval gate | Change control | Global | 2 | Organisation | r11 | Active |
| Finance tone and disclaimers | Output quality | Global | 4 | Sub-org · Corporate | r2 | Active |
| Cost ceiling per run | Cost | Global | 2 | Sub-org · Engineering | r1 | Active |
| No code execution | Safety | Global | 1 | Agent · FinOps Advisor | r1 | Active |
Showing all 12 policies.
Application order
Layers compile from the outside in. A narrower layer can tighten what a broader one allows, never loosen it — so a business unit can be stricter than its organisation, and an agent stricter again, but neither can undo a rule set above them.
Everyone in Apple Global, the role each holds here, and where they sit on the tree.
| Person | Role | Sub-organisation | Business unit | Status | Last active |
|---|---|---|---|---|---|
| AAmitamit@appleglobal.com | CFO | Corporate | Finance | Active | 2m ago |
| AArnavarnav@sourceworx.ai | CTO | Engineering | Platform | Active | 14m ago |
| FMFarzan Manzerfarzan@appleglobal.com | Org Admin | Engineering | Platform | Active | now |
| PNPriya Nairpriya@appleglobal.com | Agent Author | Engineering | AI Fabric | Active | 1h ago |
| RDRahul Desairahul@appleglobal.com | Operator | Operations | SecOps | Active | 3h ago |
| SRSneha Raosneha@appleglobal.com | Analyst | Corporate | Finance | Active | Yesterday |
| MIMeera Iyermeera@appleglobal.com | Auditor | Corporate | Risk | Active | 2d ago |
| VSVikram Singhvikram@appleglobal.com | Agent Author | Engineering | AI Fabric | Suspended | 6d ago |
| KPKaran Patelkaran@northwind.com | Guest, read only | — | — | Invited | Never |
A role is held in this organisation. The same person can hold a different role elsewhere.
How Apple Global is divided, and who sits where. Region scope is a data-residency boundary, not a label — an agent only runs where its owner is allowed.
Sub-organisations
| Sub-organisation | Regions allowed | Business units | Members |
|---|---|---|---|
| Corporate | us, eu | 2 | 11 |
| Engineering | us, eu, in | 3 | 16 |
| Operations | us | 2 | 7 |
Business units
| Business unit | Sub-organisation | Owner | Members |
|---|---|---|---|
| Finance | Corporate | Amit | 6 |
| Risk | Corporate | Meera Iyer | 5 |
| Platform | Engineering | Arnav | 7 |
| AI Fabric | Engineering | Priya Nair | 6 |
| Data | Engineering | Arnav | 3 |
| SecOps | Operations | Rahul Desai | 4 |
| Support | Operations | Rahul Desai | 3 |
People placement
| Person | Sub-organisation | Business unit | Region scope |
|---|---|---|---|
| Amit | Corporate | Finance | us, eu |
| Arnav | Engineering | Platform | us, eu, in |
| Farzan Manzer | Engineering | Platform | us, eu, in |
| Priya Nair | Engineering | AI Fabric | in |
| Rahul Desai | Operations | SecOps | us |
| Sneha Rao | Corporate | Finance | us |
| Meera Iyer | Corporate | Risk | eu |
Showing 7 of 34.
Not part of the product — this panel is here so you can test what you asked to keep changeable. Everything below changes live, across every screen above. Each palette is a complete set for both themes, so use the moon in the top bar to see its other half.
Palette
Typography
Raleway is a display face. Set both to Raleway, then flip Data face to IBM Plex Sans and watch the numeric columns in Observe — that's the tradeoff I flagged, shown rather than argued.
Colour tokens
Status colours are reserved and never reused as a chart series. Both data hues are validated for colour-blind separation against each theme's own surface — the toggle in the top bar re-steps them rather than inverting.
Every figure on this page is example data, shaped like what the real endpoints return. Nothing here is live.